{
 "schema_version": "1.0",
 "page_kind": "article",
 "pack_kind": "generate",
 "slug": "home-inference-and-diffusion-rig-cost",
 "title": "The cost to purchase and run a very capable home AI rig",
 "dek": "A consumer graphics card in a consumer desktop, nothing server-grade anywhere in the box: the rig costs $3,198.86 with one card and $5,048.85 with two on 13 September 2026, before tax. Running it costs about $18 a month with one card and $31 with two if every day puts them to work for six hours; with the cards bought used at the past year's average sale, the rig is $2,256.87 with one and $3,164.87 with two.",
 "published": "2026-09-14",
 "series": [
  "bench"
 ],
 "licence": "CC BY 4.0",
 "status": "pending-judge",
 "pack_note": "no judge is seated, so every chip is held",
 "notice": "items with published:false are held and are not this page's published words",
 "source": {
  "url": "https://research.strata2signal.com/home-inference-and-diffusion-rig-cost/",
  "md_url": "https://research.strata2signal.com/home-inference-and-diffusion-rig-cost/index.md",
  "md_sha": "f489bd4c9a6405d00fd806166dd42f64607c950967535c835f48901148214344",
  "html_sha": "0c5d32f5047f6cd2a0d1779ed3c95ef96f76bf9b163da73eaad3e61ca092de1e",
  "anchors_sha": "547ff8a7ef0ef890105c5176c0836e150ca1959af730e43133bafe91a0d899e0"
 },
 "short": {
  "paragraph": "The rig costs $3,198.86 to buy with one card and $5,048.85 with two on 13 September 2026, before tax; one card is already a very capable home AI rig, and the second is the step up to the biggest models. With the cards bought used at the past year's average sale it is $2,256.87 with one and $3,164.87 with two, or $3,537.75 with two at last month's average. Nothing in it is server-grade: two consumer cards on a consumer board, bought one part at a time. The cards are three quarters of the bill, and they moved: the pair cost $2,899.98 in May 2026 and the same listing asks $3,699.98 that morning, 28 per cent more in four months. A new card at $1,829.99 is twice the past year's average sale and 1.7 times last month's; the second-hand market rose 11 per cent over the same months. If you are buying used this month, expect about $1,100 a card; new is 1.7 times that, and every figure here is before tax and before shipping. Running the box costs far less than buying it: 91 W idle at the meter with two cards, about $12 a month. A day with both cards at their caps for six hours is about $31 a month, at the national average rate, and about $18 with one card.",
  "from_dek": false,
  "counts": {
   "words": 3553,
   "minutes": 16,
   "tables": 6,
   "kit": false
  },
  "bullets": [
   {
    "text": "the bottom tier costs $959.00",
    "figure": "$959.00",
    "cite": "where-this-rig-sits-among-the-three-tiers-and-what-is",
    "quote": "The bottom is the mini, $959.00 as one listing on 10 September 2026, which writes at about twelve tokens a second and takes minutes to read a long page.",
    "span": [
     3197,
     3205
    ]
   },
   {
    "text": "the two cards alone cost $3,699.98",
    "figure": "$3,699.98",
    "cite": "where-this-rig-sits-among-the-three-tiers-and-what-is",
    "quote": "This is the pair: the two cards alone are $3,699.98 on the same renewed listing that morning, almost four times the whole mini, before the box around them.",
    "span": [
     3450,
     3460
    ]
   },
   {
    "text": "the card reads at the wall 320 W",
    "figure": "320 W",
    "cite": "what-kind-of-performance-to-expect",
    "quote": "What kind of performance to expect One RTX 3090 on its own, from this workshop's own bench notes of 24 August 2026 (85 receipts, the model server of that day, one question at a time, measured before this workshop capped its cards), and, in the last row, the same card two days earlier serving eight questions at once through a batching runtime: words, one card reading a 26-billion-parameter mixture-of-experts writer, tokens a second 137 a dense 24-billion-parameter model, tokens a second 53 time to the first token, a 40-token prompt 0.08 s time to the first token, a 400-token prompt 0.30 s time to the first token, a 2,000-token prompt 1.01 s prompt reading, tokens a second (2,667 tokens in 0.73 s) about 3,600 the same 26-billion writer at 4-bit through a batching runtime, eight questions at once, tokens a second in aggregate (22 August 2026) 328 Pictures, one card, from the same render bench that A Rig Your Friend Already Owns published: one consumer RTX 3090, alone on its box, re-drawing this workshop's own game art on the night of 26 August 2026, 165 renders across six image models whose licences let this workshop re-share the output (five open licences and one community licence with a revenue cap), the card reading 320 W at the wall that night, above the 280 W this page recommends, so a capped card draws each of these a few per cent slower, unmeasured.",
    "span": [
     10436,
     10442
    ]
   }
  ]
 },
 "sections": [
  {
   "id": "what-this-rig-is-for-in-plain-words",
   "heading": "What this rig is for, in plain words",
   "level": 2,
   "span": [
    2018,
    2922
   ],
   "chunks": [
    [
     2018,
     2922
    ]
   ],
   "chars": 904,
   "digest": "this box runs AI models at home for inference and diffusion with no cloud or accounts. two cards provide 48 GB of graphics memory, enough for a 30-billion-parameter language model with a long page, an image model at full size, or both simultaneously.",
   "digest_skipped": null
  },
  {
   "id": "where-this-rig-sits-among-the-three-tiers-and-what-is",
   "heading": "Where this rig sits among the three tiers, and what is measured",
   "level": 2,
   "span": [
    2922,
    4021
   ],
   "chunks": [
    [
     2922,
     4021
    ]
   ],
   "chars": 1099,
   "digest": "this rig is the top of three home tiers. the bottom is $959.00 as of 10 September 2026. the middle tier is a single used card in a desktop. this tier uses a pair of cards costing $3,699.98 for the cards alone.",
   "digest_skipped": null
  },
  {
   "id": "nothing-in-this-box-is-server-grade",
   "heading": "Nothing in this box is server-grade",
   "level": 2,
   "span": [
    4021,
    4582
   ],
   "chunks": [
    [
     4021,
     4582
    ]
   ],
   "chars": 561,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "the-bill-on-one-day",
   "heading": "The bill, on one day",
   "level": 2,
   "span": [
    4582,
    6885
   ],
   "chunks": [
    [
     4582,
     6885
    ]
   ],
   "chars": 2303,
   "digest": "the rig's cost was read from amazon.com on 13 September 2026. the cards cost $1,849.99 each. other parts like memory, power supply, CPU, motherboard, storage, case, and cooler are listed as real or representative. the rig with two cards costs $5,048.85.",
   "digest_skipped": null
  },
  {
   "id": "one-card-or-two",
   "heading": "One card or two",
   "level": 2,
   "span": [
    6885,
    9028
   ],
   "chunks": [
    [
     6885,
     9028
    ]
   ],
   "chars": 2143,
   "digest": "one card provides 24 GB for a 30-billion-parameter model or full-size images. a second card adds flexibility to run two models at once, increase throughput via batching, or split one large model across both cards. the two-card rig costs $5,048.85 at the renewed EVGA listing.",
   "digest_skipped": null
  },
  {
   "id": "what-kind-of-performance-to-expect",
   "heading": "What kind of performance to expect",
   "level": 2,
   "span": [
    9028,
    12047
   ],
   "chunks": [
    [
     9028,
     12047
    ]
   ],
   "chars": 3019,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "what-the-card-is-like-for-training",
   "heading": "What the card is like for training",
   "level": 2,
   "span": [
    12047,
    14578
   ],
   "chunks": [
    [
     12047,
     14578
    ]
   ],
   "chars": 2531,
   "digest": "the card can train LoRA adapters for ACE-Step 1.5. training a music adapter for Chopin took 4 minutes 49 seconds and peaked at 5.7 gigabytes. a house-music adapter took 39 minutes 56 seconds of training.",
   "digest_skipped": null
  },
  {
   "id": "two-notes-on-the-bill",
   "heading": "Two notes on the bill",
   "level": 2,
   "span": [
    14578,
    15268
   ],
   "chunks": [
    [
     14578,
     15268
    ]
   ],
   "chars": 690,
   "digest": "the 32 GB memory is priced as two 2 × 8 GB kits because no 4 × 8 GB DDR4-3200 kit was available. the representative motherboard may run the second card on fewer lanes than the first.",
   "digest_skipped": null
  },
  {
   "id": "the-cards-in-may-and-in-september",
   "heading": "The cards, in May and in September",
   "level": 2,
   "span": [
    15268,
    16539
   ],
   "chunks": [
    [
     15268,
     16539
    ]
   ],
   "chars": 1271,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "why-these-two-are-evga-and-why-the-cheapest-card-is-a",
   "heading": "Why these two are EVGA, and why the cheapest card is a different row",
   "level": 2,
   "span": [
    16539,
    17585
   ],
   "chunks": [
    [
     16539,
     17585
    ]
   ],
   "chars": 1046,
   "digest": null,
   "digest_skipped": null
  },
  {
   "id": "what-the-rig-costs-to-run",
   "heading": "What the rig costs to run",
   "level": 2,
   "span": [
    17585,
    19748
   ],
   "chunks": [
    [
     17585,
     19748
    ]
   ],
   "chars": 2163,
   "digest": "the rig's idle draw is 91 W, costing about $12.10 a month. a working day with both cards at 280 W caps for six hours averages 233 W. the ceiling is 660 W.",
   "digest_skipped": null
  },
  {
   "id": "what-this-page-does-not-say",
   "heading": "What this page does not say",
   "level": 2,
   "span": [
    19748,
    21277
   ],
   "chunks": [
    [
     19748,
     21277
    ]
   ],
   "chars": 1529,
   "digest": "this page does not provide an asking price, only prices from one retailer on one morning. it does not benchmark the pair's word speed or provide training figures for language models.",
   "digest_skipped": null
  },
  {
   "id": "how-to-check-our-work",
   "heading": "How to check our work",
   "level": 2,
   "span": [
    21277,
    21709
   ],
   "chunks": [
    [
     21277,
     21709
    ]
   ],
   "chars": 432,
   "digest": null,
   "digest_skipped": "credits"
  },
  {
   "id": "who-ran-this-and-thanks",
   "heading": "Who ran this, and thanks",
   "level": 2,
   "span": [
    21709,
    23032
   ],
   "chunks": [
    [
     21709,
     23032
    ]
   ],
   "chars": 1323,
   "digest": null,
   "digest_skipped": "credits"
  }
 ],
 "chips": [
  {
   "id": "c-4ce7b641",
   "q": "what purpose does this specific hardware configuration serve?",
   "a": "The setup runs AI models at home, including those for answering questions and drawing pictures. It functions without sending data to external servers or requiring accounts, providing enough graphics memory to handle large language models or high-resolution image models.",
   "cites": [
    "what-this-rig-is-for-in-plain-words"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-ab0596ad",
   "q": "how does this setup compare to the lowest tier of home machines?",
   "a": "This configuration represents the top of three home tiers. The bottom tier, priced at $959.00 as of 10 September 2026, writes at approximately twelve tokens per second and requires minutes to read a long page.",
   "cites": [
    "where-this-rig-sits-among-the-three-tiers-and-what-is"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-6bdc0bea",
   "q": "what is the total cost for the full configuration?",
   "a": "The total cost for the rig with two cards was $5,048.85 on the morning of 13 September 2026. If the cards are purchased used at the past year's average sale price, the total is $3,164.87.",
   "cites": [
    "the-bill-on-one-day"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-d50a983c",
   "q": "what benefits come from using two graphics cards instead of one?",
   "a": "A second card provides flexibility by allowing a user to keep two models warm at once, run a language model and an image model simultaneously, or split one large model across both cards to fit bigger open models.",
   "cites": [
    "one-card-or-two"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-8dd3eb3f",
   "q": "how long does training a music model adapter take?",
   "a": "Training small LoRA adapters for music models takes minutes to an hour. For example, teaching a model Chopin took 4 minutes 49 seconds, while a house-music adapter took 39 minutes 56 seconds.",
   "cites": [
    "what-the-card-is-like-for-training"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-09dfb9d0",
   "q": "what are the estimated monthly electricity costs?",
   "a": "At an idle state, the cost is about $12.10 a month. A working day with both cards at their 280 W caps for six hours costs about $30.81 a month, while running at the ceiling all day costs about $87.12 a month.",
   "cites": [
    "what-the-rig-costs-to-run"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  }
 ],
 "related": [
  {
   "slug": "the-two-hour-machine-priced",
   "why": "shares ground with § What this page does not say · § What this page prices, and what it does not"
  },
  {
   "slug": "chopin-in-five-minutes",
   "why": "shares ground with § What the card is like for training · § A music model you can run at home"
  },
  {
   "slug": "half-an-hour-with-dead-composers",
   "why": "shares ground with § What the card is like for training · § The bill"
  }
 ],
 "thanks": "## Who ran this, and thanks {#who-ran-this-and-thanks}\n\nThe prices are amazon.com's own product pages on 13 September 2026 and jawa.gg's published sale averages for the RTX 3090 the same morning. The photograph is the workshop's own. The pictures beside this page's row on the shelf, and its link card, are renders based on photographs of the real rig, and nothing on the page itself is a render. A vision model on this workshop's own seat, Gemma 4 26B (Google DeepMind, Apache 2.0), described an operator's photograph of the box in one paragraph; FLUX.2 klein 4B (Black Forest Labs, Apache 2.0) painted that description in a cyberpunk world of 2055, through ComfyUI (GPL-3.0), on this very box, one of its two cards, in 2.8 to 3.2 seconds a frame at 1,344 by 768; thirteen frames were drawn across two batches and four kept, every one re-saved without metadata. The photograph itself will follow when its owner has cropped it.\n\nA small human team bought the cards, checked the prices and signed the numbers; a fleet of AI agents read the retailer, did the arithmetic and drew the pictures under that team's rulings. Neither the retailer nor the marketplace nor any model's maker owed us anything. Thanks to the readers of the mini page who asked what a real card costs; that question is the whole reason this page exists.",
 "kit": null,
 "seat_class": {
  "writer": "gemma-class",
  "judge": null,
  "writer_runtime": "vllm"
 },
 "bench": {
  "bullets_written": 3,
  "bullets_kept": 3,
  "digests_written": 12,
  "digests_kept": 8,
  "chips_written": 8,
  "chips_kept": 6,
  "chips_grounded": 0,
  "chips_published": 0,
  "dropped_by": {
   "new_noun": 2,
   "figure": 0,
   "length": 0,
   "cite": 0,
   "judge": 0,
   "quote": 0,
   "directive": 0,
   "redaction": 4,
   "profanity": 0,
   "identifier": 0
  },
  "new_noun_tokens_checked": [
   "048.85",
   "1",
   "1.5",
   "10",
   "12.10",
   "13",
   "137",
   "164.87",
   "2",
   "2026",
   "233",
   "24",
   "26-billion-parameter",
   "280",
   "3",
   "30-billion-parameter",
   "30.81",
   "3090",
   "32",
   "320",
   "39",
   "4",
   "449.99",
   "48",
   "49",
   "4B",
   "5",
   "5.7",
   "56",
   "660",
   "699.98",
   "8",
   "849.99",
   "87.12",
   "908.00",
   "91",
   "959.00",
   "ACE-Step",
   "AI",
   "CPU",
   "Chopin",
   "DDR4-3200",
   "EVGA",
   "FLUX.2",
   "GB",
   "GeForce",
   "LoRA",
   "May",
   "September",
   "Ultras",
   "W",
   "XC3"
  ],
  "new_noun_tokens_withheld": 6,
  "figure_definition": "v3",
  "drops": [
   {
    "kind": "digest",
    "reason": "redaction",
    "item": null,
    "item_chars": 259,
    "withheld": "a string carrying a fenced literal",
    "detail": ""
   },
   {
    "kind": "digest",
    "reason": "new_noun",
    "item": null,
    "item_chars": 9,
    "withheld": "a string naming something the article does not",
    "detail": ""
   },
   {
    "kind": "digest",
    "reason": "redaction",
    "item": null,
    "item_chars": 167,
    "withheld": "a string carrying a fenced literal",
    "detail": ""
   },
   {
    "kind": "digest",
    "reason": "redaction",
    "item": null,
    "item_chars": 149,
    "withheld": "a string carrying a fenced literal",
    "detail": ""
   },
   {
    "kind": "chip",
    "reason": "redaction",
    "item": null,
    "item_chars": 256,
    "withheld": "a string carrying a fenced literal",
    "detail": ""
   },
   {
    "kind": "chip",
    "reason": "new_noun",
    "item": null,
    "item_chars": 9,
    "withheld": "a string naming something the article does not",
    "detail": ""
   }
  ]
 },
 "generated_utc": "2026-09-15T13:17:35Z"
}